Caleb Hataj
// builder #28 · interviewed Oct 28, 2026

Caleb Hataj

@gearcaleeeb
A journeyman CNC machinist taking over the custom gear shop he grew up in, who replaced a 30 year old ERP with his own agent-first system, puts agents on purchasing and quoting, and builds Omarchy plugins like Detailed Weather, ThemeBook and WorkScape.
github/calebhatclaudecursoromarchy-pluginsquickshell★ Plugin author
based inWisconsin
first computerA Windows PC like everybody, and linux experiments in middle school
omarchy since2025
modelsclaude, gpt
// shorts

Shorts

16 videos
0:25When Claude got good enough to trust, it became his obsessionMy story almost exactly mirrors DHH's in his Lex Fridman podcast, where it was, like, December, January of this last year, all of a sudden, Claude...
0:273,000 slugs of bar stock in a 50-person shop, now tracked by lengthAs a, as a job shop, we, actually, just last week, we have 3,000 individual slugs of material on hand for a 50-person shop. Right? Because we...
0:43It takes 10 years to train a quoter, and machinists don't want to leave the floorI have the AI extract a list of jobs, and I just go through and look through them and mark them. Yep, this is a gold job. This one's good to...
0:43An ERP that drove him nuts, AutoHotkey, and AI as a tutorOur ERP software just drove me nuts. It's an, it's old, it's kind of, you know, it's a 30-plus years old program. And I started out just messing...
0:39Eight AI sessions in Omarchy's tiling windows, and he became the limiting factorSo I started messing with Omarchy, and I got into the window tiling manager, and I was like, holy smokes, this is sweet. So I, I had, you know...
0:2890% off a legacy ERP, pushing updates live to prodThe whole year has been a whirlwind, and we're probably 90% converted over to a fully custom ERP system away from our legacy one. Because I've...
0:31Still waiting for the day he runs out of things to buildI keep waiting for the moment when I'm gonna run out of things to do, where I don't have an agent session running all day long. It hasn't...
0:42Why he can't outsource it, and a million lines at 400 PRs a monthSo developer-wise, I think it'd be impossible for me to go to a secondary, you know, somebody, I don't know, to outsource what we've done here. I...
0:34Replacing paper with tablets on the shop floorI've got a dozen tablets in the shop for the first time. I'm testing a fully tablet inspection-type system. So we can do all of our inspection...
0:40Three pages became one operation, and why big ERPs can't do thatOur saw department, they used to have to go to one page to do their data collection, to put their time entry on there. They'd have to go to...
0:26He hardly knows what a PR is, and his agents send them for himI've used GitHub personally just for a couple private repos, but, yeah, ask me what a PR is, and I hardly know, to be honest. I don't know how to...
0:32The Omarchy plugin explosion and a renaissance in operating systemsWhen 4.0 came out, I did start making my own plugins for everything. The plugin marketplace just exploded. It was like two weeks, and there was...
0:47A self-taught solo dev who found his software world on XI've always been so isolated, right? I've always been a solo dev. I'm self-taught. I've never been part of a team working on stuff. And X has been...
0:44A weekly agent loop that trains his quoting engine on 30 years of jobsI built a whole UI where every time we have one of these prints analyzed, I can go through and add comments to every single thing that is wrong...
0:42Steel vendors toured his shop and finally saw AI making a real impactSo we had a steel company, one of our vendors, came in here a couple months ago, and they started, they just asked what we're doing for software...
0:34A four-day nitriding process they could never track, now properly quotedWe've got better cost tracking than we have ever had before. We have a heat treating operation that we do in-house called nitriding, and we've...
// life timeline
2026Claude becomes good enough to trust, and he starts replacing the shop's 30 year old ERP
2026Gets a Framework 13 and runs Omarchy on it
2026Ships Detailed Weather, ThemeBook and WorkScape after Omarchy 4.0 opens plugins
2026Puts an agent in charge of raw steel and bar stock purchasing
2026Builds a print analyzer into the ERP that improves itself every week
// 00

The rundown

Caleb Hataj is a journeyman CNC machinist who is taking over the custom gear shop he grew up in. He is also its software team. In under a year he replaced a 30 year old ERP with his own agent-first system, and he runs it from an Omarchy laptop. He does not think anyone else could have built it for him:

"I'm the domain expert. I know exactly what my company needs." ↳ jump to L70

“I'm the domain expert. I know exactly what my company needs.”Caleb Hataj · L70
// 01

Grew up on the shop floor

Caleb grew up cleaning machines on the shop floor. He started college in mechanical engineering, came back to the business, and spent his first six or seven years running machines until he finished his journeyman apprenticeship. ↳ jump to L44 The shop makes nothing but custom gears, with an average order of maybe five pieces, so some machinists do 10 to 15 setups a day. ↳ jump to L48

Everyone expected him to become a software guy. He was experimenting with linux in middle school, just because it was there. ↳ jump to L48

// 02

Thirty clicks at a time

When Caleb moved into the office to do quotes and order entry, the old ERP drove him nuts. It needed so many clicks, windows and pop ups that he taught himself AutoHotKey and automated 30 clicks at a time. Then the first AI models arrived and he used them as a tutor, because he is fully self-taught. He started pulling data out of the database to build new views, like run lists for the machinists. ↳ jump to L48

In January 2026, Claude became good enough that he could trust the code, and it became his obsession. ↳ jump to L48 He had already been using Omarchy since one of the very early betas. The tiling window manager won him over, and soon he had two monitors full of Claude sessions. "I became the limiting factor all of a sudden." ↳ jump to L52

“I became the limiting factor all of a sudden.”Caleb Hataj · L52
// 03

An ERP built for one shop

Ten months later the shop is about 90 percent moved to his custom ERP, with only invoicing and payments still being carefully converted. When someone brings him a change on the shop floor, he snaps a screenshot, tells an agent to fix it, and ships to production soon after. Every day people come to his office with new ideas, and the list never runs out. ↳ jump to L52 The app is now over a million lines of code. ↳ jump to L80

He designs every screen so older machinists can learn it too. ↳ jump to L86 The saw department used to visit three pages to do one job. Now it is one operation, and material certs follow every slice cut from a bar. ↳ jump to L86 An in-house heat treating step that sits in the oven for up to four days finally has its own data collection page, so the shop can quote it properly. ↳ jump to L282 Five other shops have seen the system, and every one of them asked if he is selling it. ↳ jump to L94 If he ever does, he would package it as a modular ERP with agents at its core. ↳ jump to L104

// 04

Agents on purchasing and quoting

His agent now orders all of the shop's raw steel and bar stock. "So I literally trained it like an employee." ↳ jump to L54 It has its own purchasing email address, sends requests for quotes to several vendors, builds its own purchase orders and picks the best quote that fits the schedule. When it needs data it cannot reach, it asks his coding agent to build the API. ↳ jump to L54 The shop tracks every bar down to one inch slugs, and the agent reads vendor PDFs and enters the data the machinists used to type by hand. ↳ jump to L128

The next step is the front office. Staff will drag a quote request into a folder, and an agent will split the attachments, create the job and fill in the customer. ↳ jump to L194 The hard part is the limits. "I'm having to be very purposeful about what I tell it not to do, almost more so than how I'm telling it what to do." ↳ jump to L194

He is also building ShopHawk, a print analyzer for quoting. It reads drawings, some hand drawn decades ago, maps the gear data into the ERP, and a weekly job turns his corrections into a pull request that improves the engine. It is trained on 30 years of the shop's own jobs. ↳ jump to L256

“So I literally trained it like an employee.”Caleb Hataj · L54
// 05

Plugins on a Framework 13

Caleb's Omarchy machine is a Framework 13. ↳ jump to L158 When Omarchy 4.0 opened plugins, he made a detailed weather plugin, a read aloud plugin for highlighting text and listening to it, ThemeBook to organize themes and WorkScape for workspace layouts. ↳ jump to L160 He builds on other people's plugins and sends fixes back through his agents. "it allows collaboration on a level that I've never seen before." ↳ jump to L168

“It's a renaissance with operating systems.”Caleb Hataj · L168
// 06

In his words

Caleb was always a solo developer, and X became his connection to the wider software world. ↳ jump to L174 "It's a renaissance with operating systems." ↳ jump to L168 He tries to sell Omarchy to nearly everyone he meets. ↳ jump to L168 When a steel vendor toured the shop, they said it was the first time they had seen AI make a real difference in a business. As Caleb puts it, "I've got it in production everywhere." ↳ jump to L278

“I try to sell Omarchy to every person I talk to, practically, these days.”Caleb Hataj · L168
// their setup

Setup

hardware
Framework 13 running Omarchydaily · laptop
tools & models
claudecursoromarchy-pluginsquickshellhyprlandgithubtailscaleclaudegptelixirqml
// full transcript (cleaned)
Show the cleaned transcript (51 lines) · open full page
L1[redacted: call setup and consent chatter, lines 1 to 43]
L44Caleb: Yeah, so I'm not actually using AI much for the CNC stuff. I'm actually a third generation, taking over the business right now. So that's how I got started with all the CNC stuff. So I kind of grew up cleaning machines on the floor. you know, working part-time, I went off to school, starting for mechanical engineering, realized I didn't want to do 7 years of math classes non-stop. So I ended up coming back to the business, and I worked on the shop floor for the first 6-7 years, got my journeyman apprenticeship done, ran.
L45[redacted: call chatter, lines 45 to 47]
L48Caleb: Yeah, what, gosh, I… that was… I'd work part-time here in the summers, working on, you know, running machines and stuff, but we're a custom job shop making nothing but gears, so… you know, some of our guys are doing 10-15 setups a day on their lathes, reprogramming, doing setups. It's not a production environment at all. Our average order size is maybe 5 pieces. So… And yeah, so that's how I kind of got into this whole side of it, but that was kind of the easy path after school, right? But going into school, everyone thought I was going to be a software guy. They thought I was super into computers. You know, I was… I was just downstairs doing, you know, hacking around and messing around with… experimenting with linux in middle school, just trying it out, because it was there, you know, why not? So, yeah, my first computer was Microsoft, like everybody. I still remember doing the keyboard classes in elementary school, and it was so new at the time. To get those keyboard classes in. So, coming back to the business. Yeah, what happened there? So I was on the shop floor forever, and then… yeah, and so then eventually I needed to transition into the office, so I started doing, quotes and order entries and, you know. the guys have to know how to make the parts on the shop first before they can quote it. So that's where I had… I had to be on the shop for a long time, I had to know, you know, ran a whole bunch of different machines, got that figured out. And then they had me start coming to the office and start doing that, and right off the bat. our ERP software just drove me nuts. It's an… it's old, it's kind of, you know, it's a 30-plus years old program. So that was 4, 5, 6 years ago now. And I started out just messing with AutoHotKey, a scripting language on Windows to control your keyboard and mouse. And the program just had… it required so much clicking and so many different windows and modals that popped up everywhere. So I just started teaching myself how to do that. And I had to mess around with Arduinos and stuff in college a little bit. I made an electric guitar with a MIDI keyboard on the back of it, just for fun. You know, I had a synthetics and composites class that I had to do that for. So I had a little bit of experience, but then I started messing with that, and that… like, instantly, it was way better. It took a bit to figure out, obviously, but… it would automate doing 30 clicks to different screens at a time, and that was kind of like, holy crap, I can… I can automate some of this stuff and make it good, right? And then, it was about that next year, the first AI models really started coming out. And I started using them just as a tutor, just as a mentor, hey, you know, I'm fully self-taught, I've got no experience, no schooling, how do I do this? How do I do that, right? And that just steamrolled, exactly, yup. And then it was pretty much… My story almost exactly mirrors DHH's in his Lex Fridman podcast, where it was, like, December… January of this last year, all of a sudden, Claude was good enough to do stuff, and I could trust the code, right? And so then… and I had… I guess I should back up. Before that, I'd already made a couple versions of our… our… what our ERP system is now. I started going into the database and extracting out different fields and making new views for the guys on the shop floor to be able to track their run list of, you know, what jobs you have to do in what order, and it was fun, it was… it was a passion of mine. It grew really quickly. But I still just hated our ERP system. It was just so old and archaic. So then once I… once Claude came out, I think it was GPT, or not GPT, it was, Opus 5.5, I think, was the first big one. I just went to town. It was… it just became my obsession, yeah. Yeah, you know what I'm talking about. It became my obsession, my passion, to…
L49[redacted: call chatter, lines 49 to 51]
L52Caleb: Okay, okay. Yeah, so… so, yeah, come January that year, or this year, gosh, that was only, like, 10, 11 months ago, I just went full bore. I was just seeing how far I could take the system, and I, I mean, at that point, I'd already been starting to mess with Omarchy as well. So, I got into Omarchy at… like, one of the very early betas, just because I hadn't fully converted to linux or anything, we still have to use Windows for our software at work and whatnot. And all of our CAD software is pretty much Windows-only right now, too. So I started messing with Omarchy, and I got into Window Tiling Manager, and I was like, holy smokes, this is sweet. So I… I had, you know, dual monitors set up with multiple… multiple tiled windows of 8 Claude sessions going at once, just really trying to lean into it, and I became the limiting factor all of a sudden. Right? My brain power just could not keep up with how many sessions I had and all the ideas I had going, It was super fun, and I just went to town. I got a ton of stuff done, and now here we are. And I transitioned between Claude, and then I went to GPT for a while. Now I'm almost ex… yeah, I bounce back and forth with the models, like everybody did, I'd imagine. Now I'm using Cursor and agents. I went from manually managing all of my tiles and sessions to now I just talk to a couple agents, and they just send me pictures of stuff as I need updates, and, you know, I dig in if I need to, and go into plan mode in a terminal window, but, Yeah, the whole world… the whole year has been a whirlwind, and now here we are 10 months later, and we're probably 90% converted over to a fully custom ERP system, away from our legacy one. The only things that we haven't fully converted yet are the financial, like, the invoicing and some ACH information, because that's probably the most complicated part of, you know, we want to make sure we get that right. Because I've very much so been in a state of, I just push updates out to prod live to everybody, and my company has become my testing grounds to live test everything, and the speed at which the agents can make improvements and fix things enables me to be able to do that. So somebody comes into the office with a change, or I'm walking around the shop floor, and I just open up the agent now, or something on my phone, I talk into it and say, hey, on this page, I take a screenshot of their screen and say, fix this, here, there, right? and an hour later, I have a push out to prod, that… they're all able to use it, and we just keep going and going, and the features that we've been able to implement now are so far beyond what we ever had with our old ERP system. And we just keep rocking it. I keep waiting for the moment when I'm gonna run out of things to do, where I don't have an agent session running all day long. And I got a coworker that I keep telling us, I'm like, I gotta be getting close. We gotta be running out of features to do, ideas, right? It hasn't happened. Every day, I come in thinking, alright, I don't have much to do today, and within the first half hour, it's… I got 8 people coming to my office with ideas and different changes, and I mean, I'm constantly running out of usage limits on a couple different accounts that I have now, because it's just… How can I not? How can I not make improvements when I can see it, and I can see the difference on the shop floor and in the office floor? So…
L53Nix: How much money would you… like, are you guys saving money?
L54Caleb: Yeah, so that's… that's the other side of it. Just within the last month, I have the agent actually ordering all of my raw steel and bar stock for the company. So I literally trained it like an employee. I set up its own purchasing email address, and… and this is the… this is the crazy part, that I think… that I'm seeing other companies really struggle with. And why I think we're getting so far ahead on this. Because our ERP system now is completely custom and Agentic-first. I'm putting in APIs for my agents. I'm having my agents build their own APIs, right? So I've got my purchasing agent. It knows where to go in my shop stock page to, you know, every time a job is released to the shop, it calc… the app itself is calculating out, okay, we got these bars on hand, we're able to track literally down to 1-inch slugs, all the way up to the 12-footers. And I'm able to exactly know how much we need based on stuff. So then I've got the agent going in there and saying, okay, here's the API for this, and then anytime I have an idea or I tell it to order something, and it can't do it, I have it set up to just go talk to my coding agent and say, hey, go build this API for me to get access to the data that I need. It's amazing, right? And now I've… now I've taken it to the level where my… Right? We used to use verbal POs for a lot of stuff, because the data entry side was just too much. As a job shop, we… actually, just last week, we have 3,000 individual slugs of material on hand for a 50-person shop. Right? Because we track down to 1-inch slugs, because we're doing so much variable lengths of materials that our saws are having to cut. that… We save the thin lengths, just in case we have a part that comes in one day, and we can use that slug. So, our old system couldn't do that. Our old system could just tell us how many feet we have on hand. Now we can track every length. But because it was verbal POs, it was… it was too much work to put in for one person. They'd have to sit there all day and create manual POs of every line item of bar, so we just sent off… we just called the person, the steel providers on the phone and say, hey, use my initials and the date for the PO. And we'll just figure it out from there. Now I've got the agent doing all the data entry. So not only is it sending out all the RFQ requests to multiple vendors, so we're getting better pricing and have more availability of most of our materials. It's also now… I had it build its own APIs to go in and create its own purchase orders with all the line items. It's tracking and updating all of the jobs that we have. It even tracks the due date for different SAW operations on jobs, and then cross-compares that against all the vendor quotes, and then picks the best quote based on which one will fit our schedule while having the best price. It used… it would take literally 2-3 people, full-time staff, to do what I have the agent doing in a day. And now we've got new dashboards, custom layouts to track our minimum stocking agreements, and… It just keeps going on and on, the amount of ridiculous stuff that we're able to automate now, and our office staff is running leaner than ever.
L55[redacted: private, lines 55 to 66]
L67Nix: So, my first thing is… What do you see about… business application development, where you had to take your idea of the app that you wanted and go, two years ago, go to a developer and try to explain what you wanted, and then they would try to convert that back and give you the application, say that ERP. But now, now the developer has come to you, if that makes sense. Instead of you going to the developer, the development has come to you, and now all you have to do is explain what you want. And that's one question. Talk about, like, how it's flipped. Instead of you going to ask for it, they do it for you. And then the other thing is, has it changed your hiring profile? Like, are you looking for somebody that kind of does the work, but also understands the philosophy of what you're doing? And then third question is. Are you sharing your ideas with other machinist companies, or are you, like… how much of a competitive edge is it giving you? And that's my three questions.
L68[redacted: call chatter, lines 68 to 69]
L70Caleb: Yeah, so, developer-wise, I think it'd be impossible. For me to go to a secondary you know, somebody, I don't know. To outsource what we've done here. I don't think it'd work, because I'm the domain expert. I know exactly what my company needs. To be able to operate at a level that it's never operated before. And I'm just the ideas guy. The, you know, the AI has enabled me to be the ideas guy. I have to get hands-on and really dive in there sometimes, but, overall, I mean, I… I now wake up in the middle of the night with an idea, and I quickly go into my phone and, like, oh, before I forget it, like, let's get this down, and then I wake up, and I have a PR ready to go with screenshots, and everything's ready to go. It's awesome! Right? And then I can just push it to prod from my house, and then by the time I get to work, the updates are in. It's awesome.
L71Nix: That's insanity.
L72Caleb: Yeah, and we run our own local server for everything, so I'm not having to worry about, you know, doing AWS or any of that.
L73[redacted: call chatter and prices, lines 73 to 79]
L80Caleb: Yeah, well… Yep, through the hiring process. Yeah, I got it up on my screen now. Okay, cool. Yeah, so, yeah, developer, and then… Well, and back to the first question, I, you know, I've asked AI every once in a while, what would it take to make my app from scratch with a traditional program, your developer suite, right? And my app's over a million lines of code by this point. I'm putting out 400 PRs a month. It's just nuts.
L81Nix: GitHubbing it? Like, a private GitHub?
L82Caleb: Yep, yep, yep. And this is… this is on top of doing all of our other job duties, on top of doing all of our quotes, and our orders, and, like, this is just agents I keep running on the side that I check in on every 10-15 minutes if there's an issue. Yeah, okay, so then the second question, the hiring process, I would say no at this point. And that is… Well… I'm purposely designing the app so that I don't have to do that. Right? I'm making the app as user-friendly as possible, because people on the shop floor have a wide range of experience with the tools. Most machine shops that we talk to have a… I mean, that's actually how we've gotten most of our work, is other machine shops having to close down their in-house building. shops, because they don't have people coming in to train and run their machines. So, I, I have…
L83Nix: Is that a competitive edge for you, that people on the floor kind of get with the technology?
L84Caleb: 100%. But you're also.
L85Nix: You're also giving… you're also giving up the experienced machinist that's like, oh, that looks like a good idea, but I'm gonna tell you why it's not. So, like, you got some expertise there, and there's a little bit of a give and take, is there not?
L86Caleb: There is, yes. It's been a struggle to get some of the guys on board. Like, actually, right now, we're… I've got a dozen tablets in the shop for the first time. I'm testing a fully tablet inspection-type system. So we can do all of our inspection data on the tablets that's saved to the jobs, we have attachments, because we've always been paper copies and handwritten notes and this and that. I'm digitifying as much as I can, where it makes sense, and some spots it doesn't yet. But I've got guys testing tablets right now for all of that. But I've still got guys running older equipment, running… you know, they're just not as tech-savvy. So I'm having to design the app in a way… that's probably been most of my efforts has been just in the UI, UX, to make it so that… because the agents have… typically been struggling with that, in my experience. They're great on the code, but, you know, it's a struggle to make it do what you want for the looks and the feel. It's getting better, but I have to design it so that those guys can be accommodated and still be able to do their job and learn the new system. But I've been able to combine so many processes that our old system kept separate, which I don't blame the old system for that. a typical developer, the amount of stuff that I'm trying to do would just be… ridiculous to try to combine all the different operations. Like, our saw department, they used to have to go to one page to do their data collection, to put their time entry on there. They'd have to go to another page to do, to process the material requirement, and then they'd have to go to a third page, to… try to do something else. Now I've got all of that in one spot as one operation, they're attaching the material certs to the jobs, and then all of that information, like, the material certs track with every single job that cuts a slice off that bar. So it's a level of data tracking that we've never been able to have before, that the old systems, it would have just been ridiculous to try to code that in. And because it's custom… I mean, every machine shop's different, right? So, we're able to put things into the software that most shops most, most ERP systems Wouldn't be able to develop because they are forced to broaden their umbrella to the widest amount of machine shops possible. But because we're able to build custom software, we're able to add all these features that the big production ERP systems, it's not worth their time to spend time on. Right? That makes sense.
L87[redacted: call chatter, lines 87 to 93]
L94Caleb: Yeah, exactly. Yeah, and that leads right into the third question. I've shown our system to 5 other shops now that have come through, just looking for other stuff, you know, looking at, hey, I want to buy a machine, whatever. Every single one of them has asked if I'm selling it yet.
L95Nix: Yeah.
L96Caleb: Because…
L97Nix: Ultimately.
L98Caleb: Yeah, they're all using outdated systems, and part of the big problem for machine shops in general We all are… we all have these old databases that have their own schemas, and to try to migrate that to a new system is a nightmare. You don't want to.
L99[redacted: call chatter, lines 99 to 103]
L104Caleb: Yeah, that's how it is. I mean, I've talked to… we got another shop that was thinking, oh, maybe we could be a trial shop to try it out, right? But they have a totally different inspections method. They're using different processes, different machines, like, and that's where every shop's gonna be different. And I've thought a lot about it. If I did decide to sell it, I'd package it as a modular, customized ERP. Where we lean into the Agentic part of it, and, you know, hey, if you've got a feature that you want, our time to delivery is going to be less than any other system that's not using AI to code it. Right? And we have to build in the blocks, and have the core architecture, and, you know, kind of make it modular.
L105Nix: Subscription, like everything else.
L106Caleb: Yeah, and I'm making the… I'm making the API integrations agent agnostic, so we can pop in anybody else if we want, but… but I don't see any other… ERP system out there leaning into, like, Agentic helpers to do all of your data entry.
L107[redacted: the host's own tools and call chatter, lines 107 to 127]
L128Caleb: I was manually… sending out RFQ request. to a couple vendors myself, and trying to… I'd have to manually print out the PDFs, or whatever, of all the quote requests, and try to compare. It's probably saved me 2-4 hours a day. of sitting there, sending out requests, processing the POs, getting the stuff updated in our system, and that's just me, and then… for the guys on the shop floor, it's every single bar that comes in, it's saving them probably 10-15 minutes, because they don't have to… they don't have to enter in the data, because most of it's already entered in there. I literally have the agent converting the PDFs to a text format it can read, and it's extracting everything, the due dates, the price, the bar names. I have it updating, creating new sizes for me if we can't find something in stock. Like…
L129Nix: That's insane.
L130Caleb: It's nuts.
L131[redacted: the host's own background, lines 131 to 154]
L155Nix: you got a Framework 13, so walk us through, your own plugins, and what do you use, and how do you like that Frame 13?
L156Caleb: Sure. Yeah, I got this…
L157[redacted: call chatter]
L158Caleb: Yeah, yeah, I got this, kind of right when Claude started coming out this beginning of this year. Cause I just needed something mobile. And I like the concept of the framework, you know, modular and, you know.
L159Nix: Yup.
L160Caleb: So plugin-wise, I don't have that much anymore. I went crazy when the plugins went out, like, that's when I made that… I made that detailed weather plugin, and I just, you know, I started customizing a whole bunch of stuff, but… I mean, looking at my plugin bar here now, I got a Tailscale, I have a text-to-voice reader plugin that I made myself. I like to listen to audiobooks while I work, so I can just highlight stuff on a page and hit super R, and it copies and starts reading it to me. Let's see here, I got… a little to-do list, a little finance one for some stocks. I made another plugin called ThemeBook that I can have, like, pre-made layouts, or the ThemeBook was actually, it's like a big theme organizer. And so I used that one quite a bit, and then I got WorkScape, so that one is another one that I made. that it… I can essentially pre-make tiling pages for different workspaces, so I just hit Fresh Workscape, and it, you know, auto-loads all my stuff.
L161Nix: So that brings me to a good question. So, I'm guessing that you're somewhat new to open source and GitHub and that whole concept?
L162Caleb: Yeah.
L163Nix: Like, new as in, like, the last year?
L164Caleb: Yeah, yeah, I've used GitHub personally just for a couple private repos.
L165Nix: Gotcha, gotcha. So, your workscape and the… they really build on other people's plugins, is that correct?
L166Caleb: Yeah, I've borrowed from a couple other people's stuff as a start, because I know that's the theme with open source. Everybody just kind of builds on each other's work, and you keep making stuff better and better. And some of the plugins that I've used I forked them off myself, and I just want something a little bit more customized, and why not? And that's the power of Omarchy that really sold me on it. What is this other plugin? I have this other one, I can't remember what it's called.
L167[redacted: call chatter]
L168Caleb: Yeah, I mean, when the… when the… when 4.0 came out. I did start making my own plugins for everything, just to cut, you know, the different ideas that I had, and it was probably… I mean, what, the plugin marketplace just exploded. It was, like, 2 weeks, and there was already a thousand plus plugins or something in there, like crazy. And then I… Then, looking around on X, I saw a lot of people coming out with similar ideas, and like, you could just tell, people were going crazy. It was awesome. It was like a new renaissance with operating… I mean, it's still going on. It's a renaissance with operating systems. I try to sell Omarchy to every person I talk to, practically, these days. Because you pair it with the agents, and it's just such a no-brainer. Why would you not want to have an operating system That you can customize to exactly what you want. And so… yeah, it was probably 3 or 4 weeks after the plugins came out, and I… I'd been blowing through my usage, my AI limits on my subscriptions, making plugins, and I was like, I need to save some of this for work. So then, by demand, I essentially said, why am I making these from scratch? There's gotta be stuff other people have already made. So then I literally started telling AI, go scan the plugin marketplace, and I started bookmarking stuff on X all the time of new plugins I saw, And then I just plug in a couple little extra data points in the background, or rearrange stuff that I want here and there, and putting together PRs for other people's repos. I can tell my agents to say, hey, I made this update to somebody's app and I found a bug on it. Go make a PR to their repo to fix it. And boom, it's done. Like, I don't even have to think about it. It just… it allows collaboration on a level that I've never seen before.
L169[redacted: call chatter, lines 169 to 173]
L174Caleb: Appreciate it. Well, it's interesting, I… I just… I've always been so isolated, right? I've always been a solo dev, I'm self-taught, I've never been part of a team working on stuff. And X has been my connection to the software world that I've never had before. I'm able to… like, I used to be on X for crypto for a while. And now it's fully transitioned to software development, Omarchy, AI usage stuff, like, it's just… it's so cool. I talk to people in real life, and they just have no idea what's going on with some of this stuff. But X has become my way to connect and share and get new ideas for a lot of stuff. And so I lean into it, because it… I feel like it's part of my job at this point to scan through X a couple times a day, just to look for new ideas for stuff that I can implement into my ERP software or my agent harness setup, and Just try to stay up with it, because the models are changing practically every week. You know, people are able to do more and more stuff. I've… I went from full terminal you know, management with sessions to… I ended up making my own plugins on Omarchy for my entire dev setup, where I could track all of my work trees and have my… you know, I could track which agents are working on which task.
L175[redacted: call chatter]
L176Caleb: Now I'm at the point where the business is… we're on pace for a record year right now.
L177[redacted: private, lines 177 to 185]
L186Caleb: Oh, yeah. Yeah, 100%, because you have to train it like an employee, especially for the purchasing and some of the other stuff that I set up. I had to train it like it was an employee. Do this in the X situation, do this in Y situation, right? And it's good enough to remember.
L187[redacted: call chatter, lines 187 to 189]
L190Caleb: Yeah, we're using our domain email accounts. I just have them… most of my employees actually don't have X accounts, so we're just registering them with the company emails. We have a cursor team account that it's all a part of now for Unified Billing. Yep, yep. And then I'm starting to get into some of the team agent account stuff, where you can have agents that have shared secrets and setup stuff. So, what I'm working on right now is one of my One of my 12 things I'm trying to set up right now, I'm trying… I'm making an agent to help my front office handle quotes and purchase requests. And so, I created… an API to… that the agents can have access to, where I can… I'm still testing this out, but I got a team agent that knows everything about our ERP system, that I can just have everybody load on, it knows how to connect to the server, get access to the API, it has all that pre-set up. So then they just have to add the team agent to their account. And then, I have another agent that knows how to manage the emails, and it knows what the ERP agent is supposed to get for, like, the data maps, to be able to load information into our system. So then the… and I haven't proved this out yet, but I know it's gonna happen, because every idea I have seems to work out with AI. It's just how it works.
L191[redacted: call chatter, lines 191 to 193]
L194Caleb: So, the goal is we have, and I… we have… dedicated folders in their Outlook accounts, they just click and drag a quote request or a purchase order into that folder. the agent checks it once an hour or whatever. And then it knows… it has all the information to do all the manual data entry, split up the PDF attachments, auto-attach them to a job skeleton, or to an RFQ skeleton, do all that initial research, fill in the customer, create a new customer in our system if it doesn't exist yet. Just save all these manual data entry steps that people have been having to do by hand. and enable the agents to just fill it in if… because it has the right API connections to know how to do it all. And the main problem I've realized is where to put the restraints, because it can do everything, right? But you've got to be able to limit it so it doesn't go crazy. And it's got access to the database, so I'm having to be very purposeful about what I tell it not to do, almost more so than how I'm telling it what to do.
L195[redacted: the host's own projects and call chatter, lines 195 to 255]
L256Caleb: Yeah, I mean, even just random, you know, I'm even… just something else cool that I'm working on, I'm even working on a print analyzing session for auto-quoting prints. Right, so as a machine shop, and this is how most machine shops run, is it's just kind of… you use your gut to quote a part. I mean, we're doing… upwards of two dozen quotes a day sometimes, completely different parts. Like, we have… we have four to five full-time office staff that do nothing but quoting and order processing for… I mean, we're… we're a… we're about a 50-person shop, and we have about 1,300 active jobs in here at a time, at any given moment. And then we have even more quotes that come in. So, I've got a process where I'm using AI as a vision system. and I've built into our ERP, it's called ShopHawk. But I've got it built into our ERP. I have a very strict set of mappings and information that our system needs to be able to create the routing and the materials, and to match it up with what we see in our print. And then I've got a built-in loop that, I can set, right? So we upload a print, and we're only doing this with stuff that's not under NDA and whatnot, right? And sometimes our prints are ancient. We got prints from the 40s that are hand-drawn that we'll have to do sometimes. And the agents have gotten good enough that they can interpret a lot of the information and extract the gear data and the dimensions off of the blueprints and whatnot. So… Vision System has… it gets handed a very specific mapping of all the data that I want to try to extract from that print. It pulls it out, and then that gets handed off to the ERP that then has this engine built in to try to map all those things up. And then I can go through and add… I built a whole UI where every time we have one of these prints analyzed, I can go through and add comments to every single thing that is wrong, everything that's right, here and there, and then I've got a once-a-week cron job that goes in. It spins up an agent to analyze everything that was wrong, everything that was right, and it makes a PR to optimize the engine for the next week. And now it's becoming this self-iterative training loop to get our quoting system to be as automated as possible. And it's probably 90% accurate to get the right routing and material to match what my guys on the shop floor are used to seeing. Because I have 30 years of training data in our database Of what our jobs are supposed to look like. Right? So why wouldn't I just… yeah, why wouldn't I just tap into all of that and use it to train all of our data? So I… I have one of my processes, as I go through a list of… I have the AI extract a list of jobs, and I just go through and look through them and mark them. Yep, this is a gold job, this one's good to analyze and use for training data. And I just keep improving that loop over and over with the goal that… because it takes 10 years to train a guy on the shop floor, to have the amount of knowledge to come into the office and quote for the amount of Versatility and expertise that we need. So I'm plugging in all that and slowly trying to bridge that gap so that I don't have to train someone for 10 years to get them off the shop floor. Most of these guys don't want to come in the office. They're shop machinists. They want to stay on the shop floor working with their hands. So it's… for some of these guys, it's like pulling teeth to get them to come in the office and run a computer all day and sit on their butt, right? It doesn't work. So I'm trying to… that's another gap that I'm trying to bridge that agents are enabling us to do at this point.
L257[redacted: call chatter and prices, lines 257 to 261]
L262Caleb: Just a small Wisconsin business, just kicking butt, putting AI into everything.
L263[redacted: call wrap up, lines 263 to 277]
L278Caleb: We had a steel company come… one of our vendors came in here a couple months ago, and they started… they just asked what we're doing for software, and I said, well, let me tell you. And they kind of made a similar comment. They're like, you know, with all the data center stuff in the news, negative or positive, and you hear all of this stuff about AI, They were excited because this… walking through our shop and seeing what we're doing, it was the first time they really saw someone actually, like, using it for something that was making a really huge impact on their business. And not just something surface level that, yeah, you're experimenting with this thing over here, and maybe it'll work, but no, I've got it in production everywhere. My whole shop is converted at this point. Every person on the shop floor now uses the new software, and like I said, we're… the only things left are some money accounting stuff that we're really taking our time with, so…
L279Nix: Are you doing, like, financial graph dashboards so you can see where your money's coming from and your average.
L280Caleb: So…
L281Nix: price, all that stuff. Yeah. That's amazing.
L282Caleb: Yep, yeah, we've got better cost tracking than we have ever had before, because I've been able… you know, even like our, we have a heat treating operation that we do in-house called nitriding. And we've never had a way to track it, because you do a little bit of labor, manual labor, you gotta clean the part, put it in the oven, and then it sits there for, you know, up to 4 days in this heat-treated oven, going through a chemical process. And I was able to make a custom data collection page just for that. That now we're able to track and actually have proper quoting for that, that we've never been able to do. So, and then that's helping the guys in the shop floor, because now they can see the data, and oh yeah, this is actually what it's costing, not just a guess. So…
L283[redacted: call wrap up, lines 283 to 304]